Field sites and climate manipulations
Four sites along a natural rainfall gradient from Northern to Southern Israel were chosen because they share most abiotic characteristics except for mean and variance in rainfall (detailed description: Supporting Information Fig. S1, and Tielbörger et al. 2014). They have Mediterranean-type climate and similar mean annual temperatures (17.7–19.1°C), yet spanned a more than eight-fold difference in mean annual rainfall (MAP). They represent mesic-Mediterranean (MM, 780 mm MAP), Mediterranean (M, 540 mm), semi-arid (SA, 300 mm) and arid conditions (A, 90 mm), respectively. The average length of the rainy season increases from the driest (Dec–March) to the wettest site (Oct–May). Rainfall variability among years decreases from the driest (40% Coefficient of Variation) to the wettest site (25% CV) (Metzet al. 2018). All sites are semi-natural shrublands with mostly annual species in the inter-shrub matrix, located at south exposures on limestone bedrock with similar elevation (470–620 m a.s.l.).
Three rainfall manipulation treatments (dry, control, wet) were imposed on entire resident plant communities for ten years (2002-2012) in the two intermediate sites, M and SA. Each treatment was replicated by five randomized 10m × 25m plots per site. The two extreme sites, A and MM, contained only five control plots each and extended the range of the natural rainfall gradient. For dry manipulation, rainout shelters intercepted 30% of each rainfall event; for wet manipulation, additional sprinkler irrigation after every other rain event added in total 30% of the site’s MAP; control plots received ambient rainfall (details in Tielbörger et al. 2014). The magnitude of the dry manipulations realistically approximated climate change scenarios for our study region, predicting 20% less rainfall until 2050 (Smiatek et al. 2011; Samuels et al . 2013), while the wet manipulations extended the general range of scenarios that were uncertain at the onset of our study.